Results 1 - 10
of
16
Workload-aware load balancing for clustered web servers
- IEEE Trans. Parallel Distrib. Syst
, 2005
"... Abstract—We focus on load balancing policies for homogeneous clustered Web servers that tune their parameters on-the-fly to adapt to changes in the arrival rates and service times of incoming requests. The proposed scheduling policy, ADAPTLOAD, monitors the incoming workload and self-adjusts its bal ..."
Abstract
-
Cited by 33 (5 self)
- Add to MetaCart
(Show Context)
Abstract—We focus on load balancing policies for homogeneous clustered Web servers that tune their parameters on-the-fly to adapt to changes in the arrival rates and service times of incoming requests. The proposed scheduling policy, ADAPTLOAD, monitors the incoming workload and self-adjusts its balancing parameters according to changes in the operational environment such as rapid fluctuations in the arrival rates or document popularity. Using actual traces from the 1998 World Cup Web site, we conduct a detailed characterization of the workload demands and demonstrate how online workload monitoring can play a significant part in meeting the performance challenges of robust policy design. We show that the proposed load balancing policy based on statistical information derived from recent workload history provides similar performance benefits as locality-aware allocation schemes, without requiring locality data. Extensive experimentation indicates that ADAPTLOAD results in an effective scheme, even when servers must support both static and dynamic Web pages. Index Terms—Clustered Web servers, self-managing clusters, load balance, locality awareness, workload characterization, static and dynamic pages. 1
Load unbalancing to improve performance under autocorrelated traffic
- In proceedings of IEEE Intl. Conf. Distributed Computing Systems (ICDCS
, 2006
"... Size-based policies have been shown to successfully balance load and improve performance in homogeneous cluster environments where a dispatcher assigns a job to a server strictly based on the job size. While the success of size-based policies is based on separating jobs to different servers accordin ..."
Abstract
-
Cited by 10 (3 self)
- Add to MetaCart
(Show Context)
Size-based policies have been shown to successfully balance load and improve performance in homogeneous cluster environments where a dispatcher assigns a job to a server strictly based on the job size. While the success of size-based policies is based on separating jobs to different servers according to their sizes by avoiding the unfavorable performance effects of having short jobs been stuck behind long jobs, we show that their effectiveness quickly deteriorates in the presence of job arrivals that are characterized by correlation in their dependence structure. We propose a new policy that still strives to separate jobs to servers according to their sizes, but this separation is biased by the effort to reduce the performance loss due to autocorrelation in the streams of jobs that are directed to each server. As a result of this effort, not all servers are equally utilized (i.e., the load in the system becomes unbalanced) but the performance benefits of this load unbalancing are significant. The proposed policy can be used on-line, i.e., it does not assume any knowledge neither of the correlation structure of the arrival stream, nor of the job size distribution in the system. Via detailed trace-driven simulation we quantify the performance benefits of the proposed policy and we show that it can effectively self adjust its configuration parameters to improve performance under continuously changing workload conditions.
Bottlenecks and their performance implications in e-commerce systems
- In WCW
, 2004
"... Abstract. We present a detailed workload characterization of a multi-tiered system that hosts an e-commerce site. Using the TPC-W workload and via experimental measurements, we illustrate how workload characteristics affect system behavior and operation, focusing on the statistical properties of dyn ..."
Abstract
-
Cited by 7 (3 self)
- Add to MetaCart
(Show Context)
Abstract. We present a detailed workload characterization of a multi-tiered system that hosts an e-commerce site. Using the TPC-W workload and via experimental measurements, we illustrate how workload characteristics affect system behavior and operation, focusing on the statistical properties of dynamic page generation. This analysis allows to identify bottlenecks and the system conditions under which there is degradation in performance. Consistent with the literature, we find that the distribution of the dynamic page generation is heavy-tailed, which is caused by the interaction of the database server with the storage system. Furthermore, by examining the queuing behavior at the database server, we present experimental evidence of the existence of statistical correlation in the distribution of dynamic page generation times, especially under high load conditions. We couple this observation with the existence (and switching) of bottlenecks in the system. Keywords: TPC-W, bottleneck identification, workload characterization, query time distribution, autocorrelation. 1
Nuccio: “Kernel-based Web switches providing content-aware routing
- IEEE NCA
, 2003
"... Locally distributed Web server systems represent a cost-effective solution to the performance problems due to high traffic volumes reaching popular Web sites. In this paper, we focus on architectures based on layer-7 Web switches because they allow a much richer set of possibilities for the Web site ..."
Abstract
-
Cited by 6 (1 self)
- Add to MetaCart
(Show Context)
Locally distributed Web server systems represent a cost-effective solution to the performance problems due to high traffic volumes reaching popular Web sites. In this paper, we focus on architectures based on layer-7 Web switches because they allow a much richer set of possibilities for the Web site architecture, at the price of a scalability much lower than that provided by a layer-4 switch. In this paper, we compare the performance of three solutions for layer-7 Web switch: a two-way application-layer architecture, a two-way kernel-based architecture, and a one-way kernel-based architecture. We show quantitatively how much better the one-way architecture performs with respect to a two-way scheme, even if implemented at the kernel level. We conclude that an accurate implementation of a layer-7 Web switch may become a viable solution to the performance requirements of the majority of cluster-based information systems. 1
Performance-Guided Load (Un)Balancing under Autocorrelated Flows
"... Abstract—Size-based policies have been shown in the literature to effectively balance load and improve performance in cluster environments. Size-based policies assign jobs to servers based on the job size and their performance improvements are an outcome of separating “short ” from “long ” jobs, by ..."
Abstract
-
Cited by 5 (3 self)
- Add to MetaCart
(Show Context)
Abstract—Size-based policies have been shown in the literature to effectively balance load and improve performance in cluster environments. Size-based policies assign jobs to servers based on the job size and their performance improvements are an outcome of separating “short ” from “long ” jobs, by avoiding having short jobs waiting behind long jobs for service. In this paper, we present evidence that performance improvements due to this separation quickly vanish if the arrival process to the cluster is autocorrelated. Based on our observations, we devise a new size-based policy called D_EQAL that still strives to separate jobs to servers according to job size but this separation is now biased by an effort to reduce performance loss due to autocorrelation in the arrival flows of jobs that are directed to each server. As a result of this bias, all servers may not be equally utilized (i.e., load in the system may be “unbalanced”), but performance benefits become significant. D_EQAL can be used on-line as it does not assume any a priori knowledge of the incoming workload. Extensive simulations show the effectiveness of D_EQAL under autocorrelated and uncorrelated arrival streams and illustrate that the policy successfully self-adjusts the degree of load unbalancing based on monitored performance measures. Index Terms—Load balancing, autocorrelated arrivals, highly variable service times, self adaptive policies. Ç 1
Trade-Offs in Designing Web Clusters
, 2002
"... this article provides a simple, but sound, formulation for examining the tradeoffs in this large design space ..."
Abstract
-
Cited by 4 (2 self)
- Add to MetaCart
this article provides a simple, but sound, formulation for examining the tradeoffs in this large design space
Balanced Overlay Networks (BON): Decentralized Load Balancing via Self-Organized Random Networks .CoRR cs.DC/0411046
, 2004
"... We introduce a new load balancing paradigm, where the instantaneous in-degree distribution of a dynamic overlay network encodes the current level of free resources at each of the server/computing nodes. The allocation of a new job is executed by sampling the in-degree of nodes via a short random wal ..."
Abstract
-
Cited by 2 (0 self)
- Add to MetaCart
(Show Context)
We introduce a new load balancing paradigm, where the instantaneous in-degree distribution of a dynamic overlay network encodes the current level of free resources at each of the server/computing nodes. The allocation of a new job is executed by sampling the in-degree of nodes via a short random walk, and assigning the job to a node with the maximum in-degree (hence, the one with the most free resources) among the sampled nodes; the reduced free resources of the node is updated by randomly deleting an appropriate number of incoming edges. Similarly, when a job is completed at a node, it updates its status by adding a desired number of incoming edges, where the other end of the edges are again picked via short random walks. We show that this mapping of free resources to the degree distribution of an overlay network is well suited to any situation where each server provides similar service such as large-scale computing (cluster and grid computing) and web content mirroring. Extensive simulation results show that our random-walk based protocol leads to an almost optimal distribution of load: the steady state topology of the overlay network approaches random d-regular graphs. Thus, for applications where resources can be adequately represented by a scalar quantity, our load balancing protocol is highly efficient and effective. Implementing the protocol directly using lightweight sockets to maintain the overlay network should involve minimal overhead, however direct connections are not required since edges in the overlay network could be “virtual ” edges that represent routes in the underlying transport layer.
Impact of Technology Trends on the Performance of Current and Future Web-based Systems
"... Abstract: The hardware technology continues to improve at a considerable rate. Besides the Moore law increments of the CPU speed, in the last years the capacity of the main memory is increasing at an even more impressive rate. One of the consequences of a continuous increment of the main memory reso ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
(Show Context)
Abstract: The hardware technology continues to improve at a considerable rate. Besides the Moore law increments of the CPU speed, in the last years the capacity of the main memory is increasing at an even more impressive rate. One of the consequences of a continuous increment of the main memory resources is the possibility of designing and implementing memory-embedded Web sites in the near future, where both the static resources and the database information is kept in the main memory of the server machines. In this paper, we evaluate the impact of memory and network technology trends on the performance of e-commerce sites that continue to be an important reference for Web-based services in terms of complexity of the hardware/software technology and in terms of performance, availability and scalability requirements. However, most of the achieved considerations can be easily extended to other Webbased services. We demonstrate through experiments on a real system how the system bottlenecks change depending on the amount of memory that is (or will be) available for storing the information of a Web site, taking or not into account the effects of a WAN. This analysis allows us to anticipate some indications about the interventions on the hardware/software components that could improve the capacity of present and future Web-based services. Keywords: Web-based services, system architecture, performance evaluation, bottleneck analysis, client/server.
STUDY OF SERVER LOAD BALANCING TECHNIQUES
"... Abstract — One of the critical scheduling problems in distributed computing environment is load balancing on a cluster of replicated servers which face a constant pressure of increased network traffic and diverse load levels. The key issue in server load balancing in a DCS is to select an effective ..."
Abstract
- Add to MetaCart
(Show Context)
Abstract — One of the critical scheduling problems in distributed computing environment is load balancing on a cluster of replicated servers which face a constant pressure of increased network traffic and diverse load levels. The key issue in server load balancing in a DCS is to select an effective load balancing scheme to distribute clients ’ requests to the servers. In this paper, we have investigated the problem of server load balancing and evaluated various server load balancing policies. We have also conducted simulation study to compare the performance of various policies. Keywords-server load balancing, admission control, stateful servers, weighted round robin, shortest queue, diffusive algorithm I.
Approved by: SYSTEM SUPPORT FOR END-TO-END PERFORMANCE MANAGEMENT
, 2007
"... To my mother and father iii ACKNOWLEDGEMENTS This thesis would not have been possible without the support and advice from many indi-viduals. I would like to express my gratitude to all of them. First and foremost, I would like to thank my advisor Karsten Schwan for his support and guidance all throu ..."
Abstract
- Add to MetaCart
(Show Context)
To my mother and father iii ACKNOWLEDGEMENTS This thesis would not have been possible without the support and advice from many indi-viduals. I would like to express my gratitude to all of them. First and foremost, I would like to thank my advisor Karsten Schwan for his support and guidance all throughput my Ph.D. years. I am grateful to him for providing lot of flexibility and resources to pursue my research interests. His tireless efforts to help his students in every possible way, inspite of his busy schedule, always amazed me. I owe every bit of my doctoral career to him. I would like to acknowledge the other members of my thesis committee. I am grateful to Dejan Milojicic for his timely advice and support. His feedbacks helped me to think critically and improve the presentation of this thesis. Prof. Mustaque Ahamad, Prof. Santosh Pande, and Prof. Calton Pu, provided insightful comments that helped me improve my work. I would also like to thank my friends at Georgia Tech for sharing so many joyous and
